Methods and computer-program products for generating grasp patterns for use by a robot

Description

TECHNICAL FIELD

[0001] The present disclosure generally relates to grasp planning in robot applications and, more particularly, to methods and computer-program products for generating robot grasp patterns using a plurality of approach rays associated with a target object.

BACKGROUND

[0002] Robots may operate within a space to perform particular tasks. For example, servant robots may be tasked with navigating within an operational space, locating objects, and manipulating objects. A robot may be commanded to find an object within the operating space, pick up the object, and move the object to a different location within the operating space. Robots are often programmed to manipulate objects quickly and in a most efficient way possible.

[0003] However, calculating an appropriate grasp posture autonomously is difficult and computationally expensive. For each grasp candidate, the grasp quality must be tested to determine how well the grasp can securely hold a target object. Not all grasp patterns will yield a successful grasp of the target object. For example, the grasp pattern may cause the joints of the robot end effector to collide with the target object, or the grip provided by the grasp pattern will not be able to hold the target object. Further, a valid arm trajectory must be performed simultaneously within the trajectory planning. Such computations may slow down the on-line processes of a robot. In some cases, grasp patterns associated with a target object may also be taught by tedious manual programming or tele-operation of the robot. This process is slow and prone to human error.

[0004] .. Additionally, uncertainty may exist during robotic manipulation of a target object. For example, there may be uncertainty as to a target object's initial pose resulting from the robot's object localization system. Such uncertainty may lead to a grasp failure. Uncertainty may also exist as to object displacement and pose resulting from the dynamics of grasping and lifting a target object with an end effector of a robot.

[0005] Accordingly, a need exists for alternative methods and computer-program products for generating successful robot grasp patterns that are developed off-line with respect to robot processes and take into consideration target object pose and displacement uncertainties.

[0006] Jean-Philippe Saut, Daniel Sidobre: "Efficient models for grasp planning with a multi-fingered hand", Robotics and Autonomous Systems, vol. 60, no. 3, pages 347-357, ISSN: 0921-8890, DOI: 10.1016/J.ROBOT.2011.07.019 presents a simple grasp planning method for a multi-fingered hand. Its purpose is to compute a context-independent and dense set or list of grasps, instead of just a small set of grasps regarded as optimal with respect to a given criterion. By context-independent, it means that only the robot hand and the object to grasp are considered. The environment and the position of the robot base with respect to the object are considered in a further stage. Such a dense set can be computed offline and then used to let the robot quickly choose a grasp adapted to a specific situation. This can be useful for manipulation planning of pick-and-place tasks. Another application is human-robot interaction when the human and robot have to hand over objects to each other. If human and robot have to work together with a predefined set of objects, grasp lists can be employed to allow a fast interaction. The proposed method uses a dense sampling of the possible hand approaches based on a simple but efficient shape feature. As this leads to many finger inverse kinematics tests, hierarchical data structures are employed to reduce the computation times. The data structures allow a fast determination of the points where the fingers can realize a contact with the object surface. The grasps are ranked according to a grasp quality criterion so that the robot will first parse the list from best to worse quality grasps, until it finds a grasp that is valid for a particular situation. Dmitry Berenson et al.: "Grasp planning in complex scenes", HUMANOIDS'07, pages 42-48, ISBN: 978-1-4244-1861-9 describes the combination of grasp analysis and manipulation planning techniques to perform fast grasp planning in complex scenes. In much previous work on grasping, the object being grasped is assumed to be the only object in the environment. Hence the grasp quality metrics and grasping strategies developed do not perform well when the object is close to obstacles and many good grasps are infeasible. We introduce a framework for finding valid grasps in cluttered environments that combines a grasp quality metric for the object with information about the local environment around the object and information about the robot's kinematics. These factors are encoded in a grasp-scoring function which is used to rank a precomputed set of grasps in terms of their appropriateness for a given scene. This ranking is essential for efficient grasp selection and experiments are presented in simulation and on the HRP2 robot. Andrew T. Miller et al.: "Automatic grasp planning using shape primitives", Proceedings of the 2003 IEEE International Conference on Robotics & Automation, vol. 2, pages 1824-1829, DOI: 10.1109/ROBOT.2003.1241860 ISBN: 978-0-7803-7736-3 notes that automatic grasp planning for robotic hands is a difficult problem because of the huge number of possible hand configurations. However, humans simplify the problem by choosing an appropriate prehensile posture appropriate for the object and task to be performed. By modelling an object as a set of shape primitives, such as spheres, cylinders, cones and boxes, a set of rules can be used to generate a set of grasp starting positions and pre-grasp shapes that can then be tested on the object model. Each grasp is tested and evaluated within a grasping simulator "GraspIt!", and the best grasps are presented to the user. The simulator can also plan grasps in a complex environment involving obstacles and the reachability constraints of a robot arm.

SUMMARY

[0007] According to the present invention, there is provided a computer-implemented method for generating grasp patterns for use by a robot as defined in appended claim 1.

[0008] According to the present invention, there is provided a computer program product as defined in appended claim 14.

[0009] These and additional features provided by the embodiments described herein will be more fully understood in view of the following detailed description, in conjunction with the drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The embodiments set forth in the drawings are illustrative and exemplary in nature and not intended to limit the subject matter defined by the claims. The following detailed description of the illustrative embodiments can be understood when read in conjunction with the following drawings, where like structure is indicated with like reference numerals and in which:

FIG. 1 depicts a schematic illustration of an exemplary robot manipulating a target object;

FIG. 2 depicts a schematic illustration of additional exemplary components of an exemplary robot according to one or more embodiments shown and described herein;

FIG. 3A depicts a schematic illustration of a target object and a plurality of approach rays associated with the target object at an approach ray density according to one or more embodiments shown and described herein;

FIG. 3B depicts a schematic illustration of the target object depicted in FIG. 3A and a plurality of approach rays associated with the target object at another approach ray density according to one or more embodiments shown and described herein;

FIG. 3C depicts a schematic illustration of the target object depicted in FIG. 3A and a plurality of approach rays associated with the target object at another approach ray density according to one or more embodiments shown and described herein;

FIG. 3D depicts a schematic illustration of the target object and a plurality of approach rays associated with a top surface of the target object depicted in FIG. 3A according to one or more embodiments shown and described herein;

FIG. 4 depicts a schematic illustration of a robot hand and a target object in accordance with a grasping pattern according to one or more embodiments shown and described herein;

FIG. 5A depicts a schematic illustration of a robot hand located at an initial position of a grasp pattern and a target object according to one or more embodiments shown and described herein;

FIG. 5B depicts a schematic illustration of a robot hand approaching a target object in accordance with a grasp pattern according to one or more embodiments shown and described herein;

FIG. 5C depicts a schematic illustration of a robot hand grasping a target object in accordance with a grasp pattern according to one or more embodiments shown and described herein;

FIGS. 6A and 6B are schematic illustrations of unsuccessful grasp patterns according to one or more embodiments shown and described herein;

FIG. 7 depicts a schematic illustration of an unsuccessful grasp pattern of a robot hand attempting to grasp a cellular phone according to one or more embodiments shown and described herein;

FIG. 8 depicts a schematic illustration of a manipulation planning module according to one or more embodiments shown and described herein; and

FIGS. 9A-9E are schematic illustrations of a robot hand grasping and lifting a target object according to one or more embodiments shown and described herein.

DETAILED DESCRIPTION

[0011] Embodiments of the present disclosure are directed to methods and computer program products for generating and filtering a grasp pattern set to develop successful individual grasp patterns for online consideration and use by a robot. More particularly, embodiments described herein reduce on-line computations by the robot in manipulating target objects by evaluating a plurality of grasp patterns associated with a variety of objects, and then providing only those grasp patterns satisfying particular criteria to the robot for the robot's consideration during on-line operation. Embodiments generate a plurality of approach rays for a target object that correspond to a plurality of possible directions that the manipulator may travel to approach and grasp the target object. Embodiments may also take into consideration uncertainties when evaluating the individual grasp patterns of the grasp pattern set, such as object pose uncertainty, and object displacement uncertainty during object manipulation. Various embodiments of methods and computer-program products for off-line generation and evaluation of robot grasp patterns are described below.

[0012] Referring initially to FIG. 1, a robot 100 according to one exemplary embodiment is illustrated. It should be understood that the robot 100 illustrated in FIG. 1 is for illustrative purposes only, and that embodiments are not limited to any particular robot configuration. The robot 100 has a humanoid appearance and is configured to operate as a service robot. For example, the robot 100 may operate to assist users in the home, in a nursing care facility, in a healthcare facility, and the like. Generally, the robot 100 comprises a head 102 with two cameras 104 that are configured to look like eyes, a locomotive base portion 106 for moving about in an operational space, a first manipulator 110, and a second manipulator 120. The first and second manipulators 110, 120 each comprise an upper arm component 112, 122, a forearm component 114, 124, and a robot hand 118, 128 (i.e., an end effector), respectively. The robot hand 118, 128 may comprise a robot hand comprising a hand portion 116, 126, a plurality of fingers joints 119, 129, and a thumb joint 119', 129' that may be opened and closed to manipulate a target object, such as a bottle 130. The upper arm component 112, 122, the forearm component 114, 124, and robot hand 118, 128 are each a particular component type of the first and second manipulator.

[0013] The robot 100 may be programmed to operate autonomously or semi-autonomously within an operational space, such as a home. In one embodiment, the robot 100 is programmed to autonomously complete tasks within the home throughout the day, while receiving audible (or electronic) commands from the user. For example, the user may speak a command to the robot 100, such as "please bring me the bottle on the table." The robot 100 may then go to the bottle 130 and complete the task. In another embodiment, the robot 100 is controlled directly by the user by a human-machine interface, such as a computer. The user may direct the robot 100 by remote control to accomplish particular tasks. For example, the user may control the robot 100 to approach a bottle 130 positioned on a table 132. The user may then instruct the robot 100 to pick up the bottle 130. The robot 100 may then develop a trajectory plan for its first and second manipulators 110, 120 to complete the task. As described in more detail below, embodiments are directed to creating trajectory plans that are optimized to provide for more human-like motion of the robot.

[0014] Referring now to FIG. 2, additional components of an exemplary robot 100 are illustrated. More particularly, FIG. 3 depicts a robot 100 and a manipulation planning module 150 (embodied as a separate computing device, an internal component of the robot 100, and/or a computer-program product comprising non-transitory computer-readable medium) for generating and evaluating grasp patterns for use by the robot 100 embodied as hardware, software, and/or firmware, according to embodiments shown and described herein. It is noted that the computer-program products and methods for generating and evaluating individual grasp patterns of a grasp pattern set may be executed by a computing device that is external to the robot 100 in some embodiments. For example, a general purpose computer (not shown) may have computer-executable instructions for evaluating individual grasp patterns. The grasp patterns that satisfy requirements of the grasp pattern evaluation may then be sent to the robot 100.

[0015] The robot 100 illustrated in FIG. 3 comprises a processor 140, input/output hardware 142, a non-transitory computer-readable medium 143 (which may store robot data/logic 144, and trajectory logic 145, for example), network interface hardware 146, and actuator drive hardware 147 to actuate the robot's manipulators (e.g., servo drive hardware). It is noted that the actuator drive hardware 147 may also include associated software to control the various actuators of the robot.

[0016] The memory component 143 may be configured as volatile and/or nonvolatile computer readable medium and, as such, may include random access memory (including SRAM, DRAM, and/or other types of random access memory), flash memory, registers, compact discs (CD), digital versatile discs (DVD), magnetic disks, and/or other types of storage components. Additionally, the memory component 143 may be configured to store, among other things, robot data/logic 144 and trajectory logic 145 (e.g., an inverse kinematic module, a pick and place planner, a collision checker, etc.). A local interface 141 is also included in FIG. 3 and may be implemented as a bus or other interface to facilitate communication among the components of the robot 100 or the computing device.

[0017] The processor 140 may include any processing component configured to receive and execute instructions (such as from the memory component 143). The input/output hardware 142 may include any hardware and/or software for providing input to the robot 100 (or computing device), such as, without limitation, a keyboard, mouse, camera, microphone, speaker, touch-screen, and/or other device for receiving, sending, and/or presenting data. The network interface hardware 146 may include any wired or wireless networking hardware, such as a modem, LAN port, wireless fidelity (Wi-Fi) card, WiMax card, mobile communications hardware, and/or other hardware for communicating with other networks and/or devices.

[0018] It should be understood that the memory component 143 may reside local to and/or remote from the robot 100 and may be configured to store one or more pieces of data for access by the robot 100 and/or other components. It should also be understood that the components illustrated in FIG. 3 are merely exemplary and are not intended to limit the scope of this disclosure. More specifically, while the components in FIG. 3 are illustrated as residing within the robot 100, this is a nonlimiting example. In some embodiments, one or more of the components may reside external to the robot 100, such as within a computing device that is communicatively coupled to one or more robots.

[0019] FIG. 2 also depicts a manipulation planning module 150 that is configured to generate a grasp pattern set, filter unsuccessful grasp patterns from the grasp pattern set, and, in some embodiments, develop manipulator and end effector motion segments to move the robot in accordance with desirable grasp patterns from the grasp pattern set. The manipulation planning module 150 is shown as external from the robot 100 in FIG. 2, and may reside in an external computing device, such as a general purpose or application specific computer. However, it should be understood that all, some, or none of the components, either software or hardware, of the manipulation planning module 150 may be provided within the robot 100. In one embodiment, grasp pattern generation, evaluation and filtering may be performed off-line and remotely from the robot 100 by a computing device such that only grasp patterns satisfying the requirements of the grasp pattern filters described herein are provided to the robot for use in manipulation planning. Further, motion planning within the manipulation planning module may also be performed off-line by an external computing device and provided to the robot 100. Alternatively, the robot 100 may determine motion planning using filtered successful grasp patterns provided by an external computing device. Components and methodologies of the manipulation planning module are described in detail below.

[0020] Referring now to FIGS. 3A-3D, a target object 130 having a plurality of approach rays associated therewith is depicted. Referring specifically to FIG. 3A, the target object 130 is configured as a box having five walls 232a-232e and an opening 233. It should be understood that the target object may be of any configuration, and embodiments are not limited to the target objects illustrated throughout the figures. As depicted in FIG. 3A, a plurality of approach rays AR are associated with the target object 130 according to a predetermined density. Each approach ray AR extends perpendicularly from a surface of the target object 130. It is noted that only a few selected approach rays AR are labeled in FIGS 3A-3D for ease of illustration.

[0021] An approach ray AR corresponds to a path of travel that the robot will traverse when approaching the target object. As described in more detail below, a single approach ray AR may have several grasp patterns associated with it. A user (e.g., a robot programmer, a robot designer, a robot operator, etc.) may change the density and the location of the approach rays AR by changing one or more settings in a user interface that is used to interact with the manipulation planning module. In this manner, users may vary the resolution of the plurality of approach rays AR about the target object. FIG. 3B depicts the same target object 130 as depicted in FIG. 3A but with a higher density of the approach rays AR. As an example and not a limitation, a user may wish to select the density setting as very small so that the grid defined by the plurality of approach rays AR is incident on a narrow object surface of the target object. However, generating a grasp pattern set based on a plurality of approach rays AR having a density that is too dense may be very time-consuming and inefficient as there may be similar successful grasp patterns analyzed during trajectory planning.

[0022] FIG. 3C depicts an approach ray density that is less than the approach ray densities of the target object depicted in FIGS. 3A and 3B. In one embodiment, a user may remove approach rays AR according certain criteria, such as approach rays having neighbors within a threshold distance may be removed. Additionally, a user may remove approach rays from a particular region or surface of the target object. FIG. 3D depicts the target object 130 wherein approach rays only extend from a top surface of walls 132a-132d. Accordingly, only grasp patterns that require the robot manipulator to approach the target object 130 from the top of the target object 130 will be considered when evaluating the grasp patterns. Use of automatic approach ray generation enables evaluation of many grasp patterns from a plurality of directions quickly and efficiently.

[0023] Referring now to FIG. 4, a schematic illustration of a robot end effector configured as a robot hand 118 approaching an exemplary target object 230 along a particular approach ray AR in accordance with an exemplary grasp pattern is provided. The robot hand 118 comprises two finger joints 119 and a thumb joint 119' that extend from a palm surface 116p of a hand portion 116. The exemplary target object includes a handle portion 231 and a base portion 239. The illustrated approach ray AR dictates that the illustrated grasp pattern will require that the robot attempt to grasp the handle portion 231 of the target object. It should be understood that only one approach ray AR is depicted in FIG. 4 for ease of illustration, and that the target object 230 may have a plurality of approach rays associated therewith.

[0024] Each individual grasp pattern of the grasp pattern set has a set of grasp parameters, which may include, but are not limited to, a pre-shape configuration of the end effector (i.e., robot hand) prior to grasping the target object, an approach ray AR of the plurality of approach rays, a standoff distance d_s, and a roll angle of the end effector.

[0025] The pre-shape configuration of the robot hand 118 is the initial finger/thumb joint 119/119' position before grasping the target object. For example, the pre-shape configuration of a particular grasp pattern may be that the finger joints 119 and the thumb joint 119' are in a fully open position. A plethora of pre-shape configurations are possible. Each grasp pattern has a single pre-shape configuration associated therewith.

[0026] As described above, each grasp pattern has a single approach ray AR. However, a single approach ray may have a plurality of grasp patterns associated therewith. The robot hand 118 (or other type of end effector) will approach the target object 230 by traversing a path dictated by the approach ray AR, commonly such that a central portion of the palm surface 116p is aligned with the approach ray AR.

[0027] The standoff distance d_s is an offset distance from a surface 232 of the target object 230 where the robot hand 118 (or other type of end effector) stops approaching the target object 230. In some grasp patterns, the standoff distance d_s may be zero such that the palm surface 116p contacts the target object 230. In other embodiments, the standoff distance d_s may be a value such that the robot hand 118 stops prior to contacting the target object, as depicted in FIG. 4.

[0028] The roll angle is the pre-rotation angle of the robot hand 118 around the approach ray AR prior to reaching the target object. The robot hand 118 may be rotated by any angle about the approach ray AR.

[0029] Other parameters may also be considered, such as a friction coefficient of the finger joints 119 and the thumb joint 119', or an amount of force exerted onto the target object by the finger joints 119 and the thumb joint 119'.

[0030] One or more combinations of the above parameters may be sent to a grasp pattern set generation module (e.g., within the manipulation planning module 150) to generate a grasp pattern set comprising a plurality of individual grasp patterns. Each approach ray AR may have a plurality of associated grasp patterns. As an example and not a limitation, an approach ray AR may be associated with a first grasp pattern resulting from a first combination of the pre-shape configuration, standoff distance, and roll angle parameters, and a second grasp pattern resulting from a second combination of the pre-shape configuration, standoff distance, and roll angle parameters. In one embodiment, the number of grasp patterns within a grasp pattern set is equal to: (the number of pre-shape configurations) × (the number of standoff distances) × (the number of roll angles) × (the number of approach arrays).

[0031] As described below, each individual grasp pattern of the grasp pattern set is evaluated by computer simulation to filter out individual grasp sets that are unsuccessful as compared to one or more grasp metrics. The grasp pattern evaluation may be performed autonomously off-line for more efficient on-line manipulation (particularly object grasping) because no evaluation of the grasp quality of various available grasp patterns is required during the on-line processes.

[0032] Each individual grasp pattern may be generated kinematically by computer simulation/calculation to determine whether or not the grasp of the individual grasp pattern is satisfactory. Referring to FIG. 5A, in one embodiment, the initial position and orientation of the robot hand coordinate system and initial finger joint angles are set before the computer simulation, which may be determined by the pre-shape configuration and roll angle parameters (as well as any other appropriate inputs or parameters). As shown in FIG. 5A, the approach ray AR of the illustrated grasp pattern dictates that the robot hand 118 will approach a top surface 234 of the handle portion 231 of the target object.

[0033] Next, the robot hand 118, by computer simulation, moves along a direction indicated by the approach ray AR until it contacts the target object or is at the standoff distance d_s. In many cases the robot hand 118 will approach the target object in a manner such that its motion is normal to its palm surface 116p, as shown by arrow B in FIG. 5B. Once the robot hand 118 is at the desired location, it may close its finger joints 119 and thumb joint 119' one by one or simultaneously until they contact the target object 230, as shown in FIG. 5C.

[0034] The contacts between the finger joints 119 and the thumb joint 119' with the target object are then extracted and used to measure a quality of the grasp. In one embodiment, a grasp quality score is generated, wherein the grasp quality score is a scalar value that indicates how securely the grasp associated with the individual grasp pattern can hold the target object. Any grasp evaluation calculation or methodology may be utilized to generate a grasp quality score. In one particular embodiment, a force-closure based measure is used as the grasp quality score for evaluating the grasp quality. A grasp may be considered a force-closure if it can resist a test force and torque without dropping the target object. The test force and torque that are applied to determine if a grasp is force-closure may depend on the target object.

[0035] The grasp quality score may then be compared with a grasp quality threshold associated with the particular target object. In one embodiment, only force-closure grasps will satisfy the grasp quality threshold. As an example and not a limitation, an individual grasp pattern wherein a finger joint does not contact the target object, or only partially contacts the target object, will have a low quality measure of the grasp. FIG. 6A, 6B and 6C depict grasp patterns that lead to unsuccessful grasps. In FIG. 6A, the finger joints 119 and thumb joint 119' strike a side surface 237 of the handle portion 231 of the target object 230 such that the robot hand 118 cannot grasp the target object 230. Similarly, as shown in FIG. 6B, one of the finger joints 119 and the thumb joint 119' strike a top surface 234 of the handle portion 231 of the target object 230. FIG. 7 depicts a robot hand 118 attempting to grasp a target object configured as a cellular phone 330 such that one of the finger joints contacts an antenna portion 331 of the cellular phone 330 without the thumb joint 119' making contact, which may lead to an unsuccessful grasp. In each of these cases, the grasp pattern associated with the grasps would be removed from the grasp pattern set. Only grasp patterns leading to force-closure grasps are provided to the robot for on-line object manipulation, according to some embodiments.

[0036] Each individual grasp pattern having a grasp quality score that is greater than the grasp quality threshold may be saved to a file, along with information regarding the individual grasp patterns. In one embodiment, each individual successful grasp pattern is finalized and saved to the file. Finalizing the grasp pattern may include, but is not limited to, saving the initial robot hand configuration before grasping (i.e., pre-shape configuration), transformation information, and the grasp quality score. The transformation information may include the relative position and orientation of the hand coordinate system with respect to the object coordinate system. The finalized individual grasp patterns may then be provided to the robot for target object manipulation (e.g., the manipulation planning module 150 may provide a file containing the finalized individual grasp patterns to the trajectory logic 145 for generation of motion segments to control the manipulator and the end effector). For example, the robot, when encountering a target object, may evaluate the provided grasp patterns when determining how to manipulate an object. As an example and not a limitation, the robot may chose the individual grasp pattern that is most closely aligned with the current position of its end effector and has the highest grasp quality score for manipulation of the target object.

[0037] In this manner, embodiments of the present disclosure filter out undesirable grasps (e.g., collision between the robot hand and the target object, non force-closure grasps, etc.), and may eliminate duplicate grasps.

[0038] Some embodiments of the present disclosure may account for uncertainties that exist when a robot attempts to grasp an object. For example, there may exist a level of uncertainty as to a target object's true pose from the robot's perspective using its vision system (e.g., a sensor based object localization system); there may be a disparity between an object's true pose and the pose detected by robot. Further, uncertainties resulting from the dynamics of an object when grasped by the robot may also be present. For example, there may be a fluctuation of a target object's pose and displacement during the grasping process.

[0039] FIG. 8 illustrates additional components of the manipulation planning module 150 according to one embodiment that provides for the inclusion of data uncertainty in estimating the grasp success rate, as well as for the incorporation of the dynamics of an object during the grasping process. At block 152, a plurality of approach arrays of a grasp pattern set is generated as described above. At block 154, the grasp quality score is calculated. In addition to the force-closure determination described above, the grasp quality score also considers the uncertainty of the target object's pose and dynamics during the grasping process via the probability distribution models of block 155, as described below. Successful individual grasp patterns are finalized and provided to the robot at block 156, while unsuccessful individual grasp patterns (e.g., non force-closure grasps, grasps that lead to significant target object slippage or displacement, and the like) are removed from the grasp pattern set, as shown in block 157. Generally, embodiments may utilize probability distribution models to account for uncertainties regarding: 1) an uncertainty as to a target object's initial pose (e.g., a pose probability distribution model); 2) an uncertainty as to an object's pose after being gripped by the robot (e.g., a reference pose probability distribution model); and 3) an uncertainty as to an objects pose or displacement after being lifted by the robot (e.g., a displacement probability distribution model).

[0040] Referring now to FIGS. 9A-9E, the grasping and lifting of a target object 400 by a robot hand 118 wherein the target object 400 is displaced is schematically illustrated. Referring initially to FIG. 9A, an uncertainty as to the actual pose of the target object 400 may be present. FIG. 9A illustrates an actual pose of the target object 400 defined by object axis OA. However, the initial pose of the target object 400 as detected by the robot may be slightly different than that illustrated in FIG. 9A. To account for this uncertainty, the manipulation planning module 150 may access a pose probability distribution model (e.g., as illustrated in block 155) to select a plurality of object poses. The probability distribution models described herein may be configured as any appropriate probability distribution model wherein the manipulation planning module 150 may select several object poses or movements from a plurality of likely object poses or movements. For example, in one embodiment, the pose probability distribution model is configured as a Gaussian probability distribution module, wherein the object pose with the greatest probability is at the peak of the Gaussian probability distribution model. Other distribution models, such as a Bayesian distribution model, for example, may be utilized.

[0041] As described above, the robot hand 118 is moved along an approach ray toward the target object 400 (FIG. 9B) by computer simulation. In FIG. 9C, the finger joints 119 and thumb joint 119' are closed about the target object 400. During the grasping process, the initial object pose of the target object 400 has changed as indicated by arrow C, resulting in a new object pose along new object axis OA'. FIG. 9D depicts that there is a displacement angle of θ between the original object axis OA and the new object axis OA'. The actual new object pose of the target object 400 may be uncertain. The manipulation planning module 150 may calculate the dynamic motion of the target object by equations of motion of the target object 400. In one embodiment, the calculation of the object pose resulting from grasping the target object 400 may also be based on a probability distribution model that incorporates the uncertainties as to the movement of the target object 400 (e.g., a reference pose probability distribution model).

[0042] Next, the robot hand 118 lifts the target object 400 by computer simulation as indicated by arrow D, and the manipulation planning module 150 calculates a relative object pose of the target object 400 resulting from the dynamics of lifting the target object 400 by the robot hand 118. The manipulation planning module 150 may also calculate a displacement of the target object 400 resulting from the lifting motion of the robot hand 118. For example, the target object 400 may slip with respect to the robot hand 118 as it is lifted, as indicated by arrow E.

[0043] A grasp quality score for each grasp pattern of the grasp pattern set may be based on several computations resulting from the sampling of one or more probability distribution models. Embodiments may also calculate the grasp quality score by comparing the pose of the target object 400 after being grasped by the robot hand 118 (i.e., a reference object pose, as shown in FIG. 9D) with the pose of the target object 400 after being lifted by the robot hand 118 (i.e., the relative object pose, as shown in FIG. 9E).

[0044] More specifically, to evaluate the quality of a single grasp resulting from a single, individual grasp pattern, multiple computations with slightly different object pose and/or displacement obtained by sampling from one or more probability distribution models are executed, so that the uncertainty of object pose and displacement may be considered. In one embodiment, multiple preliminary grasp quality scores are calculated for each sampling (or combinations of samplings) of one or more probability models corresponding to either the initial pose of the target object, or the displacement of the target object 400 during grasping or lifting. The same probability distribution model may be used for each of these purposes, or a probability distribution model for each purpose may be used. At each computation of the target object manipulation, a preliminary grasp score is generated, and the preliminary grasp score of all of the computations are averaged to determine a final grasp quality score for the particular grasp pattern.

[0045] For example, a preliminary grasp score may be generated for each initial object pose selected from the pose probability distribution model. Some of the initial object poses may lead to a force-closure grasp while some may not. Additionally, a preliminary grasp score may also be calculated for each reference object pose of the target object 400 after being grasped by the robot hand 118, and/or each relative object pose of the target object 400 after being lifted by the robot hand 118.

[0046] According to one embodiment, the following conditional logic may be applied to judge a grasp pattern as successful or unsuccessful. For each computation based on sampling from one or more probability distribution models as described above, if the target object 400 under test is out of the robot hand 118 (i.e., the robot hand 118 drops the object), or if the robot hand 118 makes contact with the target object 400 with less than two joints at the conclusion of the lift-up stage (FIG. 9E), then a minimum grasp quality score (e.g., zero (0)) may be assigned to the preliminary grasp score of the particular computation associated with the grasp pattern. On the other hand, if the displacement of the target object 400 with respect to the robot hand 118 after the lift-up stage is approximately zero, and if the robot the number of joints 119 in contact with the target object 400 after lifting the target object 400 with the robot hand 118 is greater than or equal to the predetermined contact threshold (e.g., three joints), a maximum score (e.g., one (1.0)) may be assigned to the preliminary grasp score of the particular computation associated with the grasp pattern. A median score (e.g., 0.5) may be applied to the preliminary grasp score of the particular computation associated with the grasp pattern when the number of joints 119 in contact with the target object 400 after lifting the target object 400 with the robot hand 118 is greater than or equal to the predetermined contact threshold (e.g., three joints), and the displacement of the target object 400 with respect to the robot hand 118 after lifting the target object 400 with the robot hand 118 is greater than zero and less than a predetermined displacement threshold. The preliminary grasp quality scores for the particular grasp pattern may then be averaged together and used as the grasp quality score.

[0047] In addition to the conditional logic described above, embodiments may also consider the displacement or movement of the target object under test as it is lifted during the lift-up stage. For example, a reference object pose of the target object 400 may be determined after the target object 400 is grasped by the robot hand 118 and before it is lifted (see FIG. 9D). A relative object pose of the target object 400 may also be determined after the target object 400 is lifted by the robot hand (see FIG. 9E).

[0048] Movement of the target object resulting from the lift-up stage may be calculated or otherwise determined by comparing the relative object pose to the reference object pose. For example, the maximum deviation of the relative object pose from the reference object pose during the lift-up stage may be determined. In one embodiment, if the target object 400 under test moves a significant amount during the lift-up stage, the grasp may be considered unstable, and a low grasp quality score may be given. Conversely, if there is very little movement of the target object 400 during the lift-up stage, then the grasp may be considered stable, and a highest available grasp quality score (e.g., a score of one (1.0)) may be assigned.

[0049] In one embodiment, the deviations in object position (i.e., object displacement), as determined by a relative center of mass position of the target object, and an orientation of the target object, such as object pose, are considered separately. The deviations in object position (δ_P) and object orientation (δ_R) may be computed by the following equations:



where p_com and R denote the relative center of mass position and the orientation of the target object 400 with respect to the robot hand 118, respectively. The bar on top of these parameters represents the reference values.

[0050] A grasp quality score, q, may be obtained from the maximum deviation δ^MAX of these calculations and a tolerance limit L:

If the maximum deviation for the object position or the orientation exceeds the tolerance limit, a grasp quality score of zero (0) may be given. The quality metric may be computed separately for object position and orientation and used as such, or these metrics may be combined (e.g., by averaging).

[0051] Each of the grasp quality scores described herein may be used separately, or combined to determine the final grasp quality score. As described above, individual grasp patterns of the grasp pattern set that are below a grasp quality threshold may be removed from the grasp pattern set and not provided to the robot. In this manner, the robot will only have access to successful, stable grasp patterns during on-line processes.

[0052] It should now be understood that embodiments described herein generate grasp patterns for use by a robot by generating a plurality of approach rays associated with a target object to the grasped by the robot. The density of the approach rays associated with the target object may be adjusted. Further, approach rays may be removed from particular surfaces of the target object, if desired. One or more grasp patterns are then associated with each approach ray and are evaluated by determining a grasp quality score. In some embodiments, a force-closure method is used to determine the grasp quality score. In some embodiments, probability distribution models are sampled to account for uncertainties due to object pose as well as the dynamics of the target object during grasping and lifting of the target object. In this manner, embodiments provide for a manipulation planner wherein a grasp pattern set is generated and evaluated off-line autonomously. Grasp pattern outputs are saved for use by the robot in trajectory planning for manipulation tasks. Accordingly, the complex grasp planning problem is converted into a static range searching problem, i.e., finding an appropriate grasp pattern from the plurality of successful grasp patterns.

[0053] While particular embodiments have been illustrated and described herein, it should be understood that various other changes and modifications may be made without departing from the spirit and scope of the claimed subject matter. Moreover, although various aspects of the claimed subject matter have been described herein, such aspects need not be utilized in combination. It is therefore intended that the appended claims cover all such changes and modifications that are within the scope of the claimed subject matter.

Claims

1. A computer-implemented method for generating grasp patterns for use by a robot, the method comprising:

generating, using a processor, a plurality of approach rays associated with a target object, wherein each approach ray of the plurality of approach rays extends perpendicularly from a surface of the target object;

generating, using the processor, at least one grasp pattern for each approach ray of the plurality of approach rays to generate a grasp pattern set of the target object associated with the plurality of approach rays, wherein each individual grasp pattern is based at least in part on a pre-shape configuration of a robot hand prior to a grasping motion, a standoff distance of the robot hand toward the target object, and a roll angle of the robot hand prior to the grasping motion;

calculating, using the processor, a grasp quality score for each individual grasp pattern of the grasp pattern set, wherein the grasp quality score comprises a scalar value that indicates a grasp security of the target object by the robot hand using the individual grasp patterns;

comparing, using the processor, the grasp quality score of each individual grasp pattern with a grasp quality threshold;

selecting, using the processor, individual grasp patterns of the grasp pattern set having a grasp quality score that is greater than the grasp quality threshold; and

providing, using the processor, the selected individual grasp patterns to the robot for on-line manipulation of the target object.

2. The method of claim 1, wherein each individual approach ray is associated with more than one grasp pattern.

3. The method of claim 1 or 2, wherein the plurality of approach rays is generated in accordance with a density setting.

4. The method of claim 1, 2 or 3, wherein individual grasp patterns of the grasp pattern set are generated by computer simulation.

5. The method of any preceding claim, wherein each individual grasp pattern comprises the pre-shape configuration of the robot hand of the robot, and a transformation of the robot hand with respect to the target object.

6. The method of claim 1, wherein each individual grasp pattern of the grasp pattern set is generated by:

by computer simulation, determining an initial position and an orientation of a robot hand coordinate system associated with the robot hand in accordance with the standoff distance and the roll angle according to the pre-shape configuration;

opening finger joints of the robot hand;

translating the robot hand along a selected approach ray of the plurality of approach rays until the robot hand is positioned at a predetermined distance from the target object, wherein the selected approach ray is associated with the individual grasp pattern;

closing the finger joints of the robot hand about the target object; and

determining a contact force between the finger joints of the robot hand and the target object.

7. The method of claim 6, further comprising determining whether each individual grasp pattern is a force-closure grasp based on the contact force between the finger joints of the robot hand and the target object, wherein the grasp quality score is based at least in part on the contact force.

8. The method of any preceding claim, further comprising, for each individual grasp pattern:

selecting a plurality of object poses for the target object from a pose probability distribution model;

for each selected object pose of the plurality of object poses:

by computer simulation, grasping the target object with the finger joints of the robot hand;

lifting the target object with the robot hand;

determining a number of finger joints in contact with the target object after lifting the target object with the robot hand;

determining a displacement of the target object with respect to the robot hand after lifting the target object with the robot hand; and

calculating a preliminary grasp quality score for each individual object pose of the plurality of object poses, wherein the preliminary grasp quality score is further based at least in part on the displacement of the target object with respect to the robot hand after lifting the target object with the robot hand; and

determining the grasp quality score by averaging the preliminary grasp quality scores.

9. The method of claim 8, wherein the preliminary grasp quality score is determined at least in part by:

assigning a minimum score to an individual preliminary grasp quality score when the number of finger joints in contact with the target object after lifting the target object with the robot hand is less than a predetermined contact threshold;

assigning a median score to the individual preliminary grasp quality score when the number of finger joints in contact with the target object after lifting the target object with the robot hand is greater than or equal to the predetermined contact threshold, and the displacement of the target object with respect to the robot hand after lifting the target object with the robot hand is greater than zero and less than a predetermined displacement threshold;

assigning a maximum score to the individual preliminary grasp quality score when the number of finger joints in contact with the target object after lifting the target object with the robot hand is greater than or equal to the predetermined contact threshold, and the displacement of the target object with respect to the robot hand after lifting the target object with the robot is approximately equal to zero.

10. The method of claim 8 or 9, wherein the displacement of the target object with respect to the robot hand after lifting the target object is based on a displacement probability distribution model.

11. The method of claim 8, 9 or 10, further comprising, for each selected object pose:

determining a reference object pose of the target object after grasping the target object with the finger joints of the robot hand by sampling the pose probability distribution model;

determining a relative object pose of the target object after lifting the target object with the robot hand by sampling the pose probability distribution model; and

calculating a movement of the target object based at least in part on the relative object pose and the reference object pose, wherein the preliminary grasp quality score is based at least in part on the movement of the target object, wherein the reference object pose and the relative object pose may be based at least in part on a relative center of mass position of the target object and an orientation of the target object.

12. The method of any preceding claim, further comprising, for each individual grasp pattern:

calculating a plurality of preliminary grasp quality scores by sampling a probability distribution module for a plurality of computer simulations; and

determining the grasp quality score by averaging the preliminary grasp quality scores.

13. The method of claim 1, wherein the grasp quality score is calculated at least by:

by computer simulation, closing finger joints of a robot hand of the robot about the target object;

determining a reference object pose of the target object after grasping the target object with the finger joints of the robot hand by sampling a pose probability distribution model;

lifting the target object with the robot hand;

determining a relative object pose of the target object after lifting the target object with the robot hand; and

calculating a movement of the target object based at least in part on the relative object pose and the reference object pose, wherein the grasp quality score is based at least in part on the movement of the target object.

14. A computer program product for use with a computing device to generate robot grasp patterns, the computer program product comprising:

a computer-readable medium storing computer-executable instructions for generating grasp patterns that, when executed by the computing device, cause the computing device to:

by computer simulation, carry out the method of any preceding claim for generating grasp patterns for use by a robot.

Ansprüche

1. Computerimplementiertes Verfahren zur Erzeugung von Greifmustern zur Verwendung durch einen Roboter, wobei das Verfahren die Schritte umfasst:

ein Erzeugen, unter Verwendung eines Prozessors, einer Vielzahl von Annäherungsstrahlen, die mit einem Zielobjekt verknüpft sind, wobei sich jeder Annäherungsstrahl der Vielzahl von Annäherungsstrahlen senkrecht von einer Oberfläche des Zielobjektes her ausdehnt;

ein Erzeugen, unter Verwendung des Prozessors, zumindest eines Greifmusters für jeden Annäherungsstrahl der Vielzahl von Annäherungsstrahlen, zur Erzeugung eines Greifmustersatzes des Zielobjektes, das mit der Vielzahl von Annäherungsstrahlen verknüpft ist, wobei jedes einzelne Greifmuster zumindest teilweise auf einer Vorform-Konfiguration einer Roboterhand vor einer Greifbewegung, einem Distanzabstand der Roboterhand in Richtung des Zielobjektes, und einem Rollwinkel der Roboterhand vor der Greifbewegung basiert;

ein Berechnen, unter Verwendung des Prozessors, einer Greifqualitätswertung für jedes einzelne Greifmuster des Greifmustersatzes, wobei die Greifqualitätswertung einen skalaren Wert aufweist, der eine Greifsicherheit des Zielobjektes durch die Roboterhand unter Verwendung der einzelnen Greifmuster anzeigt;

ein Vergleichen, unter Verwendung des Prozessors, der Greifqualitätswertung jedes einzelnen Greifmusters mit einem Greifqualitätsschwellwert;

ein Auswählen, unter Verwendung des Prozessors, einzelner Greifmuster aus dem Greifmustersatz mit einer größeren Greifqualitätswertung als der Greifqualitätsschwellwert; und

ein Bereitstellen, unter Verwendung des Prozessors, der ausgewählten einzelnen Greifmuster an den Roboter zur Online-Manipulation des Zielobjektes.

2. Verfahren nach Anspruch 1, wobei jeder einzelne Annäherungsstrahl mit mehr als einem Greifmuster verknüpft wird.

3. Verfahren nach Anspruch 1 oder 2, wobei die Vielzahl von Annäherungsstrahlen in Übereinstimmung mit einer Dichteeinstellung erzeugt wird.

4. Verfahren nach Anspruch 1, 2 oder 3, wobei einzelne Greifmuster der Greifmustersatz durch eine Computersimulation erzeugt werden.

5. Verfahren nach einem der vorstehenden Ansprüche, wobei jedes einzelne Greifmuster die Vorform-Konfiguration der Roboterhand des Roboters und eine Transformation der Roboterhand hinsichtlich des Zielobjektes umfasst.

6. Verfahren nach Anspruch 1, wobei jedes einzelne Greifmuster des Greifmustersatzes erzeugt wird durch:

ein Bestimmen, mittels einer Computersimulation, einer Anfangsposition und einer Orientierung eines Roboterhandkoordinatensystems das mit der Roboterhand in Übereinstimmung mit dem Distanzabstand und dem Rollwinkel, entsprechend der Vorform-Konfiguration verknüpft ist;

ein Öffnen der Fingergelenke der Roboterhand;

ein Verschieben der Roboterhand entlang einem ausgewählten Annäherungsstrahl der Vielzahl von Annäherungsstrahlen, bis die Roboterhand an einem vorbestimmten Abstand von dem Zielobjekt positioniert ist, wobei der ausgewählte Annäherungsstrahl mit dem einzelnen Greifmuster verknüpft ist;

ein Schließen der Fingergelenke der Roboterhand um das Zielobjekt; und

ein Bestimmen einer Kontaktkraft zwischen den Fingergelenken der Roboterhand und dem Zielobjekt.

7. Verfahren nach Anspruch 6, ferner mit einem Bestimmen, auf der Grundlage der Kontaktkraft zwischen den Fingergelenken der Roboterhand und dem Zielobjekt, ob jedes einzelne Greifmuster ein Kraftschlussgriff ist, wobei die Greifqualitätswertung zumindest teilweise auf der Kontaktkraft basiert.

8. Verfahren nach einem der vorstehenden Ansprüche, ferner mit, für jedes einzelne Greifmuster:

einem Auswählen einer Vielzahl von Objekthaltungen für das Zielobjekt aus einem Haltungswahrscheinlichkeitsverteilungsmodell;

für jede ausgewählte Objekthaltung der Vielzahl von Objekthaltungen:

einem Greifen, mittels einer Computersimulation, des Zielobjektes mit den Fingergelenken der Roboterhand;

Anheben des Zielobjektes mit der Roboterhand;

einem Bestimmen einer Anzahl von Fingergelenken, die nach dem Anheben des Zielobjektes mit der Roboterhand mit dem Zielobjekt in Kontakt stehen;

einem Bestimmen einer Verschiebung des Zielobjektes hinsichtlich der Roboterhand nach dem Anheben des Zielobjektes mit der Roboterhand; und

einem Berechnen einer vorläufigen Greifqualitätswertung für jede einzelne Objekthaltung der Vielzahl von Objekthaltungen, wobei die vorläufige Greifqualitätswertung ferner zumindest teilweise auf der Verschiebung des Zielobjektes hinsichtlich der Roboterhand nach dem Anheben des Zielobjektes mit der Roboterhand basiert ist; und

einem Bestimmen der Greifqualitätswertung durch Mitteln der vorläufigen Greifqualitätswertungen.

9. Verfahren nach Anspruch 8, wobei die vorläufige Greifqualitätswertung zumindest teilweise bestimmt ist durch:

ein Zuordnen einer Minimalwertung zu einer einzelnen vorläufigen Greifqualitätswertung, falls die Anzahl von Fingergelenken, die nach dem Anheben des Zielobjektes mit der Roboterhand mit dem Zielobjekt in Kontakt stehen, kleiner als ein vorbestimmter Kontaktschwellwert ist;

ein Zuordnen einer Medianwertung zu der einzelnen vorläufigen Greifqualitätswertung, falls die Anzahl von Fingergelenken, die nach dem Anheben des Zielobjektes mit der Roboterhand mit dem Zielobjekt in Kontakt stehen, größer oder gleich dem vorbestimmten Kontaktschwellwert ist, und die Verschiebung des Zielobjektes hinsichtlich der Roboterhand nach dem Anheben des Zielobjektes mit der Roboterhand größer als null und kleiner als ein vorbestimmter Verschiebungsschwellwert ist; und

ein Zuordnen einer Maximalwertung zu der einzelnen vorläufigen Greifqualitätswertung, falls die Anzahl von Fingergelenken, die nach dem Anheben des Zielobjektes mit der Roboterhand mit dem Zielobjekt in Kontakt stehen, größer oder gleich dem vorbestimmten Kontaktschwellwert ist, und die Verschiebung des Zielobjektes hinsichtlich der Roboterhand nach dem Anheben des Zielobjektes mit dem Roboter ungefähr gleich null ist.

10. Verfahren nach Anspruch 8 oder 9, wobei die Verschiebung des Zielobjektes hinsichtlich der Roboterhand nach dem Anheben des Zielobjektes auf einem Verschiebungswahrscheinlichkeitsverteilungsmodell basiert.

11. Verfahren nach Anspruch 8, 9 oder 10, ferner mit, für jede ausgewählte Objekthaltung:

einem Bestimmen einer Bezugsobjekthaltung des Zielobjektes nach dem Greifen des Zielobjektes mit den Fingergelenken der Roboterhand durch Abtasten des Haltungswahrscheinlichkeitsverteilungsmodells;

einem Bestimmen einer Relativobjekthaltung des Zielobjektes nach dem Anheben des Zielobjektes mit der Roboterhand durch Abtasten des Haltungswahrscheinlichkeitsverteilungsmodells; und

einem Berechnen einer Bewegung des Zielobjektes zumindest teilweise auf der Grundlage der Relativobjekthaltung und der Bezugsobjekthaltung, wobei die vorläufige Greifqualitätswertung zumindest teilweise auf der Bewegung des Zielobjektes basiert, wobei die Bezugsobjekthaltung und die Relativobjekthaltung zumindest teilweise auf einer relativen Schwerpunktposition des Zielobjektes und einer Orientierung des Zielobjektes basieren.

12. Verfahren nach einem der vorstehenden Ansprüche, ferner mit, für jedes einzelne Greifmuster:

Berechnen einer Vielzahl von vorläufigen Greifqualitätswertungen durch Abtasten eines Wahrscheinlichkeitsverteilungsmoduls für eine Vielzahl von Computersimulationen; und

Bestimmen der Greifqualitätswertung durch Mitteln der vorläufigen Greifqua litätswertungen.

13. Verfahren nach Anspruch 1, wobei die Greifqualitätswertung durch zumindest:

ein Schließen, mittels einer Computersimulation, von Fingergelenken einer Roboterhand des Roboters um das Zielobjekt;

ein Bestimmen einer Bezugsobjekthaltung des Zielobjektes nach Greifen des Zielobjektes mit den Fingergelenken der Roboterhand durch Abtasten eines Haltungswahrscheinlichkeitsverteilungsmodells;

ein Anheben des Zielobjektes mit der Roboterhand;

ein Bestimmen einer Relativobjekthaltung des Zielobjektes nach dem Anheben des Zielobjektes mit der Roboterhand; und

ein Berechnen einer Bewegung des Zielobjektes zumindest teilweise auf der Grundlage der Relativobjekthaltung und der Bezugsobjekthaltung, wobei die Greifqualitätswertung zumindest teilweise auf der Bewegung des Zielobjektes basiert,

berechnet ist.

14. Computerprogrammprodukt zur Verwendung mit einer Computervorrichtung zur Erzeugung von Robotergreifmustern, wobei das Computerprogrammprodukt aufweist:

ein Computer-lesbares Medium, das Computer-ausführbare Anweisungen zur Erzeugung von Greifmustern speichert, die bei Ausführung durch die Computervorrichtung die Computervorrichtung veranlassen:

das Verfahren mittels einer Computersimulation nach einem der vorstehenden Ansprüche zur Erzeugung von Greifmustern zur Verwendung durch einen Roboter durchzuführen.

Revendications

1. Procédé mis en oeuvre par ordinateur pour produire des modèles de prise destinés à une utilisation par un robot, le procédé comprenant les étapes suivantes :

produire, à l'aide d'un processeur, une pluralité de demi-droites d'approche associées à un objet cible, dans lequel chaque demi-droite d'approche de la pluralité de demi-droites d'approche s'étend perpendiculairement à une surface de l'objet cible ;

produire, à l'aide du processeur, au moins un modèle de prise pour chaque demi-droite d'approche de la pluralité de demi-droites d'approche pour produire un ensemble de modèles de prise de l'objet cible associé à la pluralité de demi-droites d'approche, dans lequel chaque modèle individuel de prise est fondé au moins en partie sur une configuration de forme initiale de main de robot, avant un mouvement de préhension, une distance de sécurité de la main de robot par rapport à l'objet cible, et un angle d'inclinaison de la main de robot avant le mouvement de préhension ;

calculer, à l'aide du processeur, un score de qualité de prise de chaque modèle individuel de prise de l'ensemble de modèles de prise, dans lequel le score de qualité de prise comprend une valeur scalaire qui indique une sécurité de prise de l'objet cible par la main de robot à l'aide des modèles individuels de prise ;

comparer, à l'aide du processeur, le score de qualité de prise de chaque modèle individuel de prise avec un seuil de qualité de prise ;

sélectionner, à l'aide du processeur, des modèles individuels de prise de l'ensemble de modèles de prise ayant un score de qualité de prise qui est supérieur au seuil de qualité de prise ; et

fournir, à l'aide du processeur, les modèles individuels de prise sélectionnés au robot aux fins d'une manipulation en direct de l'objet cible.

2. Procédé selon la revendication 1, dans lequel chaque demi-droite individuelle d'approche est associée à plus d'un modèle de prise.

3. Procédé selon la revendication 1 ou 2, dans lequel la pluralité de demi-droites d'approche est produite en fonction d'un paramètre de densité.

4. Procédé selon la revendication 1, 2 ou 3, dans lequel des modèles individuels de prise de l'ensemble de modèles de prise sont produits par simulation sur ordinateur.

5. Procédé selon l'une quelconque des revendications précédentes, dans lequel chaque modèle individuel de prise comprend la configuration de forme initiale de la main de robot du robot, et une transformation de la main du robot par rapport à l'objet cible.

6. Procédé selon la revendication 1, dans lequel chaque modèle individuel de prise de l'ensemble de modèles de prise est produit par :

détermination, au moyen d'une simulation sur ordinateur, d'une position initiale et d'une orientation d'un système de coordonnées de main de robot associé à la main de robot, conformément à la distance de sécurité et à l'angle d'inclinaison selon la configuration de forme initiale ;

ouverture d'articulations de doigts de la main de robot ;

translation de la main de robot le long d'une demi-droite d'approche sélectionnée de la pluralité de demi-droites d'approche, jusqu'à ce que la main de robot soit positionnée à une distance prédéterminée de l'objet cible, dans lequel la demi-droite d'approche sélectionnée est associée au modèle individuel de prise ;

fermeture des articulations de doigts de la main de robot autour de l'objet cible ; et

détermination d'une force de contact entre les articulations de doigts de la main de robot et l'objet cible.

7. Procédé selon la revendication 6, comprenant en outre la détermination de ce que chaque modèle individuel de prise est ou non une prise à fermeture de force, sur la base de la force de contact entre les articulations de doigts de la main de robot et l'objet cible, dans lequel le score de qualité de prise est fondé au moins en partie sur la force de contact.

8. Procédé selon l'une quelconque des revendications précédentes, comprenant en outre, pour chaque modèle individuel de prise :

la sélection d'une pluralité de poses d'objet de l'objet cible d'après un modèle de distribution de probabilités de pose ;

pour chaque pose d'objet sélectionnée parmi la pluralité de poses d'objet, les étapes suivantes :

par simulation sur ordinateur, saisir l'objet cible avec les articulations de doigts de la main de robot ;

soulever l'objet cible avec la main de robot ;

déterminer un nombre d'articulations de doigts en contact avec l'objet cible, après avoir soulevé l'objet cible avec la main de robot ;

déterminer un déplacement de l'objet cible par rapport à la main de robot, après avoir soulevé l'objet cible avec la main de robot ; et

calculer un score préliminaire de qualité de prise pour chaque pose individuelle d'objet de la pluralité de poses d'objet, dans lequel le score préliminaire de qualité de prise est fondé en outre, au moins en partie, sur le déplacement de l'objet cible par rapport à la main de robot, après le soulèvement de l'objet cible avec la main de robot ; et

la détermination du score de qualité de prise en calculant la moyenne des scores préliminaires de qualité de prise.

9. Procédé selon la revendication 8, dans lequel le score préliminaire de qualité de prise est déterminé au moins en partie par :

attribution d'un score minimum à un score préliminaire individuel de qualité de prise, lorsque le nombre d'articulations de doigts en contact avec l'objet cible, après le soulèvement de l'objet cible avec la main de robot, est inférieur à seuil prédéterminé de contact ;

attribution d'un score médian au score préliminaire individuel de qualité de prise, lorsque le nombre d'articulations de doigts en contact avec l'objet cible, après le soulèvement de l'objet cible avec la main de robot, est supérieur ou égal au seuil prédéterminé de contact, et que le déplacement de l'objet cible par rapport à la main de robot, après le soulèvement de l'objet cible avec la main de robot, est supérieur à zéro et inférieur à un seuil prédéterminé de déplacement ;

attribution d'un score maximum au score préliminaire individuel de qualité de prise, lorsque le nombre d'articulations de doigts en contact avec l'objet cible, après le soulèvement de l'objet cible avec la main de robot, est supérieur ou égal au seuil prédéterminé de contact, et que le déplacement de l'objet cible par rapport à la main de robot, après le soulèvement de l'objet cible avec le robot, est sensiblement égal à zéro.

10. Procédé selon la revendication 8 ou 9, dans lequel le déplacement de l'objet cible par rapport à la main de robot, après le soulèvement de l'objet cible, est fondé sur un modèle de distribution de probabilités de déplacement.

11. Procédé selon la revendication 8, 9 ou 10, comprenant en outre pour chaque pose d'objet sélectionnée, les étapes suivantes :

déterminer une pose d'objet de référence de l'objet cible, après avoir saisi l'objet cible avec les articulations de doigts de la main de robot, en échantillonnant le modèle de distribution de probabilités de pose ;

déterminer une pose d'objet relative de l'objet cible, après avoir soulevé l'objet cible avec la main de robot, en échantillonnant le modèle de distribution de probabilités de pose ; et

calculer un mouvement de l'objet cible, sur la base, au moins en partie, de la pose d'objet relative et de la pose d'objet de référence, dans lequel le score préliminaire de qualité de prise est fondé au moins en partie sur le mouvement de l'objet cible, dans lequel la pose d'objet de référence et la pose d'objet relative peuvent être fondées, au moins en partie, sur un centre relatif de position de masse de l'objet cible et une orientation de l'objet cible.

12. Procédé selon l'une quelconque des revendications précédentes, comprenant en outre, pour chaque modèle individuel de prise, les étapes suivantes :

calculer une pluralité de scores préliminaires de qualité de prise en échantillonnant un modèle de distribution de probabilités d'une pluralité de simulations sur ordinateur ; et

déterminer le score de qualité de prise en calculant la moyenne des scores préliminaires de qualité de prise.

13. Procédé selon la revendication 1, dans lequel le score de qualité de prise est calculé au moins en partie par :

fermeture, par simulation sur ordinateur, des articulations de doigts d'une main de robot du robot autour de l'objet cible ;

détermination d'une pose d'objet de référence de l'objet cible, après la prise de l'objet cible avec les articulations de doigts de la main de robot, par échantillonnage d'un modèle de distribution de probabilités de pose ;

soulèvement de l'objet cible avec la main de robot ;

détermination d'une pose d'objet relative de l'objet cible, après le soulèvement de l'objet cible avec la main de robot ; et

calcul d'un mouvement de l'objet cible fondé au moins en partie sur la pose d'objet relative et la pose d'objet de référence, dans lequel le score de qualité de prise est fondé au moins en partie sur le mouvement de l'objet cible.

14. Produit-programme informatique destiné à une utilisation avec un dispositif informatique pour produire des modèles de prise de robot, le produit-programme informatique comprenant :

un support lisible par ordinateur stockant des instructions exécutables par ordinateur destinées à produire des modèles de prise, qui, lorsqu'elles sont exécutées par le dispositif informatique, font que le dispositif informatique :

mette en oeuvre, par simulation sur ordinateur, le procédé selon l'une quelconque des revendications précédentes pour produire des modèles de prise destinés à une utilisation par un robot.

Drawing